2021 FDA Science Forum
Datamining Medication Error Reports in the FDA Adverse Event Reporting System (FAERS) to Supplement Strategies for Identifying Potential Safety Signals
- Authors:
- Center:
-
Contributing OfficeCenter for Drug Evaluation and Research
Abstract
Background
Medication errors are preventable events that cost the American healthcare system approximately $21 billion per year according to the Network for Excellence in Health Innovation. CDER’s Division of Medication Error Prevention and Analysis (DMEPA) is responsible for surveilling and preventing medication errors. Datamining is widely used throughout FDA for detecting adverse event safety signals; however, the utility of datamining to detect medication error safety signals is unknown.
Purpose
To assess if FDA datamining strategies can detect known medication error safety signals
Methodology
Twenty-seven medication error signals, reviewed by DMEPA between 2019 and 2020, were selected for evaluation. For each signal, the drug product names, medication error events, and dates that the medication error signal was first identified by DMEPA were extracted. The drug names and events were input into FDA’s datamining software, Empirica Signal, which applied a Multi-item Gamma Poisson Shrinker algorithm to calculate the lower 5% confidence interval (EB05) of the Empirical Bayes Geometric Mean. Quarterly EB05 results for each drug-event pair were recorded back to at least 1 year before each respective signal was first identified. Success in this study was determined by a signal’s associated drug-event pair having an EB05 value of 2 or greater occurring before the quarter the safety signal was first identified.
Results
The 27 medication error signals evaluated by DMEPA corresponded to 32 drug products and 120 medication error events. Of the 27 signals, datamining successfully identified drug-event pairs associated with 13 known medication error signals before the date the error was first identified by DMEPA. The remaining 14 signals were either identified by datamining during the quarter or after the date they were first identified or were not identified due to EB05 results lower than 2.
Conclusion
Datamining successfully identified 13 of 27 medication error signals before the date they were first identified by DMEPA, highlighting the potential value that such methods could provide as a tool for signal detection. Further study involving other methods to validate datamining of medication errors is currently undergoing to strengthen these conclusions.
